Search

The remarkable increase in published medical imaging datasets for chest X-rays has significantly improved the performance of deep learning techniques to classify lung diseases efficiently. However, large datasets require special arrangements to make them suitable, accessible, and practically usable in remote clinics and emergency rooms. Additionally, it increases the computational time and image-processing complexity. This study investigates the efficiency of converting the 2D chest X-ray into one-dimensional texture representation data using descriptive statistics and local binary patterns, enabling the use of feed-forward neural networks to efficiently classify lung diseases within a short time and with cost effectiveness. This method bridges diagnostic gaps in healthcare services and improves patient outcomes in remote hospitals and emergency rooms. It also could reinforce the crucial role of technology in advancing healthcare. Utilizing the Guangzhou and PA datasets, our one-dimensional texture representation achieved 99% accuracy with a training time of 10.85 s and 0.19 s for testing. In the PA dataset, it achieved 96% accuracy with a training time of 38.14 s and a testing time of 0.17 s, outperforming EfficientNet, EfficientNet-V2-Small, and MobileNet-V3-Small. Therefore, this study suggests that the dimensional texture representation is fast and effective for lung disease classification.

Real-life time series datasets exhibit complications that hinder the study of time series forecasting (TSF). These datasets inherently exhibit non-stationarity as their distributions vary over time. Furthermore, the intricate inter- and intra-series relationships among data points pose challenges for modeling. Many existing TSF models overlook one or both of these issues, resulting in inaccurate forecasts. This study proposes a novel TSF model designed to address the challenges posed by real-life data, delivering accurate forecasts in both multivariate and univariate settings. First, we propose methods termed “weak-stationarizing” and “non-stationarity restoring” to mitigate distributional shift. These methods enable the removal and restoration of non-stationary components from individual data points as needed. Second, we utilize the spectral decomposition of weak-stationary time series to extract informative features for forecasting. To learn features from the spectral decomposition of weak-stationary time series, we exploit a mixer architecture to find inter- and intra-series dependencies from the unraveled representation of the overall time series. To ensure the efficacy of our model, we conduct comparative evaluations against state-of-the-art models using six real-world datasets spanning diverse fields. Across each dataset, our model consistently outperforms or yields comparable results to existing models.

Texture describes the unique features of an image. Therefore, texture classification is a crucial task in computer vision. Various CNN-based deep learning methods have been developed to classify textures. During training, the deep-learning model undergoes an end-to-end procedure of learning features from low to high levels. Most CNN architectures depend on high-level features for the final classification. Hence, other low- and mid-level information was not prioritized for the final classification. However, in the case of texture classification, it is essential to determine detailed feature information within the pattern to classify textures as they have diversity and irregularity in images within the same class. Therefore, the feature information at the low- and mid-levels can also provide meaningful information to distinguish the classes. In this study, we introduce a CNN model with a feature retention module (FRM) to utilize features from numerous levels. FRM maintains the texture information extracted at each level and extracts feature information through filters of various sizes. We used three texture datasets to evaluate the proposed model combined with the FRM. The experimental results showed that learning using different levels of features together assists in improving learning performance more than learning using high-level features.

In polyp segmentation, the latest notable topic revolves around polyp generalization, which aims to develop deep learning-based models capable of learning from single or multiple source domains and applying this knowledge to unseen datasets. A significant challenge in real-world clinical settings is the suboptimal performance of generalized models due to domain shift. Convolutional neural networks (CNNs) are often biased towards low-level features, such as style features, impacting generalization. Despite attempts to mitigate this bias using data augmentation techniques, learning model-agnostic and class-specific feature representations remains complex. Previous methods have employed image-level transformations with styles to supplement training data diversity. However, these approaches face limitations in ensuring style diversity due to restricted style sources, limiting the utilization of the potential style space. To address this, we propose a straightforward yet effective style conversion and generation module integrated into the UNet model. This module transfers diverse yet plausible style features to the original training data at the feature-level space, ensuring that generated styles align closely with the original data. Our method demonstrates superior performance in single-domain generalization tasks across five datasets compared to prior methods.

Technological advancements have shifted human living and working environments from outdoor to indoor. Although indoor spaces offer protection from unfavorable weather conditions, they also present new health challenges. Stale, humid, and warm indoor air creates an ideal breeding ground for bacteria and fungi, leading to health issues such as asthma and bacterial infections. Although proper ventilation is crucial, a comprehensive inspection of local indoor air quality is necessary to prevent widespread diseases. In vitro experiments involving bacteria and fungi collected from indoor air yield accurate results but are time- and cost-intensive. In silico methods offer faster results and provide valuable insights for guiding further in vitro experiments. In this study, we conduct an in vitro cytotoxicity assay on 32 fungi species and compare its results with a memory-efficient in silico modeling method using parameter-efficient fine-tuning (PEFT) and ProtBERT. This study suggests a potential methodology for predicting the toxicity of indoor airborne fungi when their identities are known.

Traditional encoder–decoder networks like U-Net have been extensively used for polyp segmentation. However, such networks have demonstrated limitations in explicitly modeling long-range dependencies. In such networks, local patterns are emphasized over the global context, as each convolutional kernel focuses on only a local subset of pixels in the entire image. Several recent transformer-based networks have been shown to overcome such limitations. Such networks encode long-range dependencies using self-attention methods and thus learn highly expressive representations. However, due to the computational complexity of modeling the whole image, self-attention is expensive to compute, as there is a quadratic increment in cost with the increase in pixels in the image. Thus, patch embedding has been utilized, which groups small regions of the image into single input features. Nevertheless, these transformers still lack inductive bias, even with the image as a 1D sequence of visual tokens. This results in the inability to generalize to local contexts due to limited low-level features. We introduce a hybrid transformer combined with a convolutional mixing network to overcome computational and long-range dependency issues. A pretrained transformer network is introduced as a feature-extracting encoder, and a mixing module network (MMNet) is introduced to capture the long-range dependencies with a reduced computational cost. Precisely, in the mixing module network, we use depth-wise and 1 × 1 convolution to model long-range dependencies to establish spatial and cross-channel correlation, respectively. The proposed approach is evaluated qualitatively and quantitatively on five challenging polyp datasets across six metrics. Our MMNet outperforms the previous best polyp segmentation methods.

Several pathogens that spread through the air are highly contagious, and related infectious diseases are more easily transmitted through airborne transmission under indoor conditions, as observed during the COVID-19 pandemic. Indoor air contaminated by microorganisms, including viruses, bacteria, and fungi, or by derived pathogenic substances, can endanger human health. Thus, identifying and analyzing the potential pathogens residing in the air are crucial to preventing disease and maintaining indoor air quality. Here, we applied deep learning technology to analyze and predict the toxicity of bacteria in indoor air. We trained the ProtBert model on toxic bacterial and virulence factor proteins and applied them to predict the potential toxicity of some bacterial species by analyzing their protein sequences. The results reflect the results of the in vitro analysis of their toxicity in human cells. The in silico-based simulation and the obtained results demonstrated that it is plausible to find possible toxic sequences in unknown protein sequences.

Recently, there have been many types of research applying drones with a thermal camera to detect deteriorations in photovoltaic (PV) modules. A thermal camera can measure temperatures on the surface of PV modules and find the deteriorated area. However, a thermal camera generally has a lower resolution than a visible camera because of the limitations of cost. Due to different resolutions between the visible and thermal cameras, there are often invalid frames from a thermal camera. In this paper, we describe a gimbal controller with a real-time image processing algorithm to control the angle of the camera to position the region of interest (ROI) in the center of target PV modules to solve this problem. We derived the horizontal angle and vertical position of ROI in visible images using image processing algorithms such as the Hough transform. These values are converted into a PID control signal for controlling the gimbal. This process makes the thermal camera capture the effective area of target PV modules. Finally, experimental results showed that the photovoltaic module’s control area was properly located at the center of the thermal image.

Several factors cause the output degradation of the photovoltaic (PV) module. The main affecting elements are the higher PV module temperature, the shaded cell, the shortened or conducting bypass diodes, and the soiled and degraded PV array. In this paper, we introduce an image processing technique that automatically identifies the module generating the hot spots in the solar module. In order to extract feature points, we used the maximally stable extremal regions (MSER) method, which derives the area of interest by using the inrange function, using the blue color of the PV module. We propose an effective matching method for feature points and a homography translation technique. The temperature data derivation method and the normal/ abnormal decision method are described in order to enhance the performance. The effectiveness of the proposed system was evaluated through experiments. Finally, a thermal image analysis of approximately 240 modules was confirmed to be 97% consistent with the visual evaluation in the experimental results.

In the last few decades, photovoltaic (PV) power station installations have surged across the globe. The output efficiency of these stations deteriorates with the passage of time due to multiple factors such as hotspots, shaded cell or module, short-circuited bypass diodes, etc. Traditionally, technicians inspect each solar panel in a PV power station using infrared thermography to ensure consistent output efficiency. With the advancement of drone technology, researchers have proposed to use drones equipped with thermal cameras for PV power station monitoring. However, most of these drone-based approaches require technicians to manually control the drone which in itself is a cumbersome task in the case of large PV power stations. To tackle this issue, this study presents an autonomous drone-based solution. The drone is mounted with both RGB (Red, Green, Blue) and thermal cameras. The proposed system can automatically detect and estimate the exact location of faulty PV modules among hundreds or thousands of PV modules in the power station. In addition, we propose an automatic drone flight path planning algorithm which eliminates the requirement of manual drone control. The system also utilizes an image processing algorithm to process RGB and thermal images for fault detection. The system was evaluated on a 1-MW solar power plant located in Suncheon, South Korea. The experimental results demonstrate the effectiveness of our solution.

Colonoscopy, which refers to the endoscopic examination of colon using a camera, is considered as the most effective method for diagnosis of colorectal cancer. Colonoscopy is performed by a medical doctor who visually inspects one’s colon to find protruding or cancerous polyps. In some situations, these polyps are difficult to find by the human eye, which may lead to a misdiagnosis. In recent years, deep learning has revolutionized the field of computer vision due to its exemplary performance. This study proposes a Convolutional Neural Network (CNN) architecture for classifying colonoscopy images as normal, adenomatous polyps, and adenocarcinoma. The main objective of this study is to aid medical practitioners in the correct diagnosis of colorectal cancer. Our proposed CNN architecture consists of 43 convolutional layers and one fully-connected layer. We trained and evaluated our proposed network architecture on the colonoscopy image dataset with 410 test subjects provided by Gachon University Hospital. Our experimental results showed an accuracy of 94.39% over 410 test subjects.

Previous studies have revealed the occurrence of alterations of white matter (WM) and grey matter (GM) microstructures in Alzheimer’s disease (AD) and their prodromal state amnestic mild cognitive impairment (MCI). In general, these alterations can be studied comprehensively by modeling the brain as a complex network, which describes many important topological properties, such as the small-world property, modularity, and efficiency. In this study, we systematically investigated white matter abnormalities using unbiased whole brain network analysis. We compared regional and network related WM features between groups of 19 AD and 25 MCI patients and 22 healthy controls (HC) using tract-based spatial statistics (TBSS), network based statistics (NBS) and graph theoretical analysis. We did not find significant differences in fractional anisotropy (FA) between two groups on TBSS analysis. However, observable alterations were noticed at a network level. Brain network measures such as global efficiency and small world properties were low in AD patients compared to HCs.

The importance of vacant parking space detection systems is increasing dramatically as the avoidance of traffic congestion and the time-consuming process of searching an empty parking space is a crucial problem for drivers in urban centers. However, the existing parking space occupancy detection systems are either hardware expensive or not well-generalized for varying images captured from different camera views. As a solution, we take advantage of an affordable visual detection method that is made possible by the fact that camera monitoring is already available in the majority of parking areas. However, the current problem is a challenging vision task because of outdoor lighting variation, perspective distortion, occlusions, different camera viewpoints, and the changes due to the various seasons of the year. To overcome these obstacles, we propose an approach based on Dilated Convolutional Neural Network specifically designed for detecting parking space occupancy in a parking lot, given only an image of a single parking spot as input. To evaluate our method and allow its comparison with previous strategies, we trained and tested it on well-known publicly available datasets, PKLot and CNRPark + EXT. In these datasets, the parking lot images are already labeled, and therefore, we did not need to label them manually. The proposed method shows more reliability than prior works especially when we test it on a completely different subset of images. Considering that in previous studies the performance of the methods was compared with well-known architecture—AlexNet, which shows a highly promising achievement, we also assessed our model in comparison with AlexNet. Our investigations showed that, in comparison with previous approaches, for the task of classifying given parking spaces as vacant or occupied, the proposed approach is more robust, stable, and well-generalized for unseen images captured from completely different camera viewpoints, which has strong indications that it would generalize effectively to other parking lots.

Electric power line equipment such as insulators, cut-out-switches, and lightning-arresters play important roles in ensuring a safe and uninterrupted power supply. Unfortunately, their continuous exposure to rugged environmental conditions may cause physical or electrical defects in them which may lead to the failure to the electrical system. In this paper, we present an automatic real-time electrical equipment detection and defect analysis system. Unlike previous handcrafted feature-based approaches, the proposed system utilizes a Convolutional Neural Network (CNN)-based equipment detection framework, making it possible to detect 17 different types of powerline insulators in a highly cluttered environment. We also propose a novel rotation normalization and ellipse detection method that play vital roles in the defect analysis process. Finally, we present a novel defect analyzer that is capable of detecting gunshot defects occurring in electrical equipment. The proposed system uses two cameras; a low-resolution camera that detects insulators from long-shot images, and a high-resolution camera which captures close-shot images of the equipment at high-resolution that helps for effective defect analysis. We demonstrate the performances of the proposed real-time equipment detection with up to 93% recall with 92% precision, and defect analysis system with up to 98% accuracy, on a large evaluation dataset. Experimental results show that the proposed system achieves state-of-the-art performance in automatic powerline equipment inspection.